Google just announced Veo 2 which produces another advance in video quality, out-performing even OpenAI’s Sora. They also announced Imagen 3, an upgraded image model also offering state-of-the-art quality.
While video models frequently “hallucinate” unwanted details—such as extra fingers or unexpected objects—Veo 2 minimizes these occurrences, resulting in more realistic outputs.
Additionally, Veo 2 embeds an invisible SynthID watermark in its videos, allowing them to be identified as AI-generated. This helps mitigate risks of misinformation and misattribution.
VisitGoogle Labs to sign up for the waitlist. They also plan to expand Veo 2 to YouTube Shorts and other products next year.
Read more about it at https://blog.google/
Imagen 3 outperformed all models, including Midjourney, Flux, and Ideogram, in human evaluations for preference, visual quality, and prompt adherence. The model is now available through Google Labs’ ImageFX.
